Download Does the real GDP per capita convergence hold in the - Hal-SHS

Does the real GDP per capita convergence hold in the
Common Market for Eastern and Southern Africa?
Amélie Charles, Olivier Darné, Jean-François Hoarau
To cite this version:
Amélie Charles, Olivier Darné, Jean-François Hoarau. Does the real GDP per capita convergence hold in the Common Market for Eastern and Southern Africa?. 2009. <hal-00422522>
HAL Id: hal-00422522
https://hal.archives-ouvertes.fr/hal-00422522
Submitted on 7 Oct 2009
HAL is a multi-disciplinary open access
archive for the deposit and dissemination of scientific research documents, whether they are published or not. The documents may come from
teaching and research institutions in France or
abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est
destinée au dépôt et à la diffusion de documents
scientifiques de niveau recherche, publiés ou non,
émanant des établissements d’enseignement et de
recherche français ou étrangers, des laboratoires
publics ou privés.
Working Paper
Document de Travail
EA 4272
Does the real GDP per capita
convergence hold in the Common
Market for Eastern and Southern
Africa?
Amélie CHARLES (*)
Olivier DARNE (**)
Jean-François HOARAU (***)
2009/25
(*) Audencia Nantes, School of Management
(**) LEMNA, Université de Nantes
(***) CERESUR, University of La Réunion
Laboratoire d’Economie et de Management Nantes-Atlantique
Université de Nantes
Chemin de la Censive du Tertre – BP 52231
44322 Nantes cedex 3 – France
www.univ-nantes.fr/iemn-iae/recherche
Tél. +33 (0)2 40 14 17 19 – Fax +33 (0)2 40 14 17 49
Does the real GDP per capita convergence hold in
the Common Market for Eastern and Southern
Africa?
Amélie CHARLES
Audencia Nantes, School of Management
Olivier DARNÉ
LEMNA, University of Nantes
Jean-François HOARAU∗
CERESUR, University of La Réunion
Abstract
This article examines the convergence of real GDP per capita in the Common
Market for Eastern and Southern Africa (COMESA) during the period 19502003. Income departures across countries were evaluated from several panel data
unit root tests, especially we consider the absolute and conditional convergence.
We find no evidence supporting the existence of convergence process for the income in the COMESA. Nevertheless, applying economic development criterion
allows to identity two absolute convergence clubs into the COMESA, one for the
most four developed countries (Egypt, Libya, Mauritius, Seychelles), and one
other for the fourteen less developed ones. Thus, we show that most economies
of COMESA are locked into a sustained poverty trap process.
Keywords:
Regional integration; convergence; Eastern and Southern Africa;
panel unit root tests.
JEL Classification: F15; O40; C12; C23.
∗ Corresponding
author: [email protected]. Address: Université de La Réunion, Faculté de
Droit et d’Économie, 15 Avenue René Cassin, BP 7151, 97715 Saint Denis Mess cedex 9.
1
1
Introduction
Testing real income convergence, i.e. convergence in per capita output across different
economies, remains one of the most challenges in the contemporaneous international
economic literature (Islam, 2003). On the whole, there are at least three main reasons
that justify the interest of study this subject. Firstly, this exercise can help to discriminate between economic growth models. On the one hand, the neoclassical model
predicts that per capita output will converge to each country’s steady-state or to a common steady-state, regardless of its initial per capita output level (Solow, 1956). On the
other hand, endogenous growth models, by underlining the importance of initial conditions and the possibility of multiple equilibriums, show that there is no tendency for
income levels to converge in the long-run (Romer, 1986, 1990). Secondly, as a consequence of the above remark, whether or not the exogenous or the endogenous version
is validated induces a potential for state intervention in the growth process. Thirdly,
on the empirical side, strong differences have been observed in per capita output and
in growth rates across countries during the last three decades, and especially between
many African economies and emerging Asian and developed economies (Maddison,
2001).
Moreover, the wave of regionalism in the 1990s has spurred academic and professional interest towards the economic effects of regional integration agreements [hereafter, RIAs]. Among these effects, a RIA is expected to strengthen trade links and
hence to facilitate technological spillovers across borders. Then, income levels should
converge and the initially poorer member states will catch up with the richer ones.
However, in a recent theoretical article, Venables (2003) states that income dispersion
across countries in a RIA will decrease only in the case of North-North integration (or
at most North-South). On the contrary, South-South integration could easily lead to
income divergence and unequal distribution of welfare gains.
Since the pioneer work of Baumol (1986) and Barro and Sala-i-Martin (1991,
1992), the test of the convergence hypothesis has consisted of fitting cross-country
regressions. Convergence is said to occur if a negative correlation is found between
the average growth rate and the initial income. However, Quah (1993, 1996) criticizes
cross-country growth regression and shows that in order to evaluate the convergence
hypothesis one must exploit the time series properties of the cross-country variances.
Moreover, Bernard and Durlauf (1996) demonstrate that the cross-section growth re-
2
gressions cannot discriminate between the hypotheses of global or local convergence.
Then, Bernard and Durlauf (1995, 1996) propose to considering convergence as a
stochastic process, using the properties of time series, and test the convergence hypothesis from unit root tests. However, time-series unit root testing has been often
criticized for its limited power and poor size properties (Haldrup and Jansson, 2006).
The small number of observations available on the time-series dimension would then
make the country-by-country analysis of income convergence in RIAs of recent formation particularly problematic. Therefore, Evans (1996) suggests exploiting both the
time-series and the cross-section information included in the data of the per capita
income in order to evaluate the convergence hypothesis. With this approach, the crosssectional and time-series information are combined, thus inducing a significant improvement in terms of power of the test.
Only few studies (McCoskey, 2002; Paap et al., 2005; Carmignani, 2006; Cuñado
and Pérez de Gracia, 2006; Guetat and Serranito, 2007; Carmignani, 2007) have been
conducted to examine convergence in African countries and, in particular, in Eastern
and Southern African economies. Therefore, this paper aims at pursuing investigations
about economic growth convergence for the main RIA of Eastern and Southern Africa,
namely the « COmmon Market of Eastern and Southern Africa » [COMESA] but
in an original way. We apply various panel unit root tests to real GDP per capita
data for 20 Eastern and Southern African countries (Angola, Burundi, Comoros,
Democratic Republic of Congo, Djibouti, Egypt, Eritrea, Ethiopia, Kenya, Libya,
Madagascar, Malawi, Mauritius, Rwanda, Seychelles, Sudan, Swaziland, Tanzania,
Uganda, Zambia, Zimbabwe): first generation tests based on the assumption of
independent cross-section units (Levin et al., 2002; Im et al., 2003); and second
generation tests allowing for cross-section dependence (Bai and Ng, 2004).
More precisely, two main issues are investigated: (1) is there an intra-regional
convergence process?, i.e. relative to the average income level of the area, among
COMESA’s members and (2) if not, are there any convergence clubs within the
COMESA? To this end three main criteria were used to test for convergence clubs,
namely
(i) the degree of human and economic development,
(ii) the membership to another regional trade agreement in Africa, and
(iii) the nature of the export base (oil producers versus non-oil producers).
3
Note that empirical testing of the convergence hypothesis provides several definitions of convergence, and thus different methodologies to test it 1 . In the convergence
debate, two definitions have emerged: the absolute convergence and the conditional
convergence. The former occurs when the level of per capita income of the poor countries catch-up with the one of the rich ones. This can be achieved if the growth rates of
developing countries are significantly higher than those of developed countries. The
latter implies that each country is converging to its own steady state and that in the
long run all the growth rates will be equalized. We consider both the absolute and conditional convergence with panel unit root tests. The absolute convergence hypothesis
uses a panel unit roots test with no fixed individual effects, whereas the conditional
convergence requires panel unit root tests with fixed individual effects.
The remainder of the paper is organized as follows. Section 2 proposes a survey
of the recent empirical works dealing with real income convergence in Eastern and
Southern African countries.
Section 3 briefly displays the econometric strategy
retained and the convergence hypothesis considered, and describes the panel unit root
tests. Section 4 presents the data and the main findings. Finally, Section 5 concludes.
2
Brief literature survey
The COMESA is a regional integration grouping of African states (Angola, Burundi,
Comoros, Democratic Republic of Congo, Djibouti, Egypt, Eritrea, Ethiopia, Kenya,
Libya, Madagascar, Malawi, Mauritius, Rwanda, Seychelles, Sudan, Swaziland,
Tanzania, Uganda, Zambia and Zimbabwe) which have agreed to promote regional
integration through trade development and to develop their natural and human
resources for the mutual benefit of all their peoples.
One of the six objectives
of COMESA as enshrined in the COMESA Treaty is to contribute towards the
establishment of the African Economic Treaty.
1 See
2
Islam (2003) for a survey on the different definitions and methodologies relative to the concept
of convergence.
2 The five others objectives is to to create and maintain: (i) a full free trade area guaranteeing the
free movement of goods and services produced within COMESA and the removal of all tariffs and
non-tariff barriers; (ii) a customs union under which goods and services imported from non-COMESA
countries will attract an agreed single tariff in all COMESA states; (iii) free movement of capital and
investment supported by the adoption of a common investment area so as to create a more favorable
investment climate for the COMESA region; (iv) a gradual establishment of a payment union based on the
COMESA Clearing House and the eventual establishment of a common monetary union with a common
currency; and (v) the adoption of common visa arrangements, including the right of establishment leading
4
COMESA was initially established in 1981 as the Preferential Trade Area [PTA]
for Eastern and Southern Africa, within the framework of the Organisation of African
Unity’s Lagos Plan of Action and the Final Act of Lagos. The PTA was transformed
into COMESA in 1994. The PTA was established to take advantage of a larger market
size, to share the region’s common heritage and destiny and to allow greater social
and economic co-operation, with the ultimate objective being to create an economic
community.
The empirical literature highlights many works which focus on the problem of
the economic growth process in Africa (e.g., Easterly and Levine, 1997; Bloom and
Sachs, 1998; Collier and Gunning, 1999; Block, 2001; Bertocchi and Canova, 2002).
However, little attention has been paid to the real convergence process both among the
countries within the African continent and with respect to developed countries. On this
subject, five papers (McCoskey, 2002; Paap et al., 2005; Carmignani, 2006; Cuñado
and Pérez de Gracia, 2006; Carmignani, 2007) must be presented.
Firstly, McCoskey (2002) investigates the convergence properties of six indicators
of well being
3
for 37 Sub-Saharan African countries. Using of both the panel unit
root test of Im et al. (2003) and the panel cointegration test of McCoskey and Kao
(1998), applied to pair-wise income differentials, McCoskey finds no evidence of
time series convergence across the whole sample for the real GDP-based variables.
Moreover, this finding still holds even for more homogeneous groups of economies
sharing some institutional arrangements such as the Southern African Development
Community (SADC) and the Southern African Customs Union (SACU)4 .
Paap et al. (2005) address the question whether or not sub-Saharan African
countries have lower average growth rates in real GDP per capita than countries
in Asia, Latin America and the Middle East over the period 1960-2000. To this
regard, they propose a latent-class panel time series model, which allows a data-based
classification of countries into clusters such that, within a cluster, countries have the
same average growth rate. Then, three clusters or three convergence clubs can be put
forward, and many Eastern and Southern African countries belong to the low growth
eventually to the free movement of bona fide persons.
3 These indicators are (i) the government share of GDP measured in 1985 international prices, (ii)
the capital stock per worker, (iii) a measurement of exports added to imports as a fraction of GDP (all
measured in current prices) (iv) a measure of real GDP per capita at 1985 international prices, (v) a
measurement of consumption added to government expenditure as a % of GDP and (vi) a measure of real
GDP per worker at 1985 international prices.
4 See Section 4 for a brief description of these RIAs.
5
cluster. Only Egypt, Mauritius, Malawi, Seychelles and Zimbabwe can be assigned to
the middle growth class and none belong to the high growth cluster.
Carmignani (2006) focuses on the problem of macroeconomic convergence for the
COMESA. The author analyzes the hypothesis of real income convergence, among
others5 , using data covering the period 1960-2002. Two measures of convergence
based on cross-country regression are computed. The first one, i.e. the so-called σ
convergence, is the standard deviation of per capita real GDP across member states.
The second one, i.e. the so-called β convergence, is the estimated coefficient on initial
(or lagged) per capita GDP in a regression of the rate of per capita GDP growth.
Carmignani concludes that income does not appear to converge across COMESA
member states. On the contrary, the gap between poorer and richer countries in the
region is widening and overall distribution is probably evolving towards a bi-modal
configuration.
In a more general article, Cuñado and Pérez de Gracia (2006) apply time series
tests to analyze both the stochastic and β-convergence conditions of per capita output
of 43 African countries to an average of the African countries and with respect to
the US economy using data for the period 1950-1999. If we just consider the results
for Eastern and Southern African area, this work finds the evidence of conditional
convergence only for the case of Seychelles towards the US economy. When the
catch-up hypothesis is retained, i.e. by taking into account a time trend when testing
the unit root hypothesis, more evidence of convergence towards the African average
(Djibouti, Egypt, Kenya, Uganda and Zimbabwe) and towards the US economy
(Egypt, Mauritius, and Seychelles) is found.
Finally, Carmignani (2007) investigates the extent of per capita income convergence in regional integration initiatives. To this end, panel unit root testing, developed
by Im et al. (2003), is performed on 28 regional groupings among which several
agreements of Eastern and Southern Africa (CBI, COMESA, SACU, SADC6 ). On the
whole, it appears that per capita income convergence is not necessarily a prerogative
of North-North integration. This hypothesis holds also for several South-South initiatives. However, this optimistic remark on the convergence properties of South-South
integration needs to be qualified. In some cases, cross-country convergence appears to
be taking place around a relatively flat regional growth trend. That is, while countries
in some South-South RIAs do converge towards the regional average, this regional av5 The author studies the degree of convergence of macroeconomic policy across members and the issue
of whether COMESA is an optimal currency area.
6 See Section 4 for a brief description of these RIAs.
6
erage fails to catch-up with industrial countries’ income. Conversely, there are RIAs
whose average income is catching-up with industrial economies, but member states
fail to converge to the regional mean. All in all, the conclusion of this paper is that
South-South integration does not necessarily imply widening intra-regional disparities.
However, it might lead to a form of convergence to the bottom.
3
The econometric strategy in a panel data framework
Nowadays, the increasing application of the panel data techniques to the determination of time-series stochastic properties has led to the development of a wide range
of new proposals in the econometric literature. The combination of the information
in the time and cross-section dimensions to compose a panel data set of individuals,
i.e. countries or regions, onto which performs the analysis of the stochastic properties
has revealed as a promising way to increase the power of these tests. The emergence
of new econometric methods has led economists to focus on the convergence debate
(Gaulier, Hurlin and Jean-Pierre, 1999; Carmignani, 2007; Guetat and Serranito, 2007;
Lima and Resende, 2007).
3.1
The income convergence hypothesis: absolute versus conditional
convergence
Several researchers have focused on the definition of the convergence concept in a
stochastic framework (e.g., Carlino and Mills, 1993; Bernard and Durlauf, 1996;
Evans, 1996; Evans and Karras, 1996; Guetat and Serranito, 2007). Islam (2003)
showed that this definition is relatively unambiguous for a two-economy situation.
However, things are different when convergence is considered in a sample of more
than two economies. Then, some have based their analysis of convergence on deviations from a reference economy although others have opted for deviations from the
sample average. Following the work of Evans and Karras (1996) and Guetat and Serranito (2007), we choose the second viewpoint.
Consider a sample of economies 1, 2, . . . , N that have access to the same body of
technological knowledge. For each economy, the convergence hypothesis implies that
a unique steady state exists, that any deviation of the state variables from their long run
values is temporary, and hence that initial values of the state variables have no effects
7
on their long run levels. The common technical knowledge assumption further implies
that the balanced growth paths of the N economies are parallel: the state variables can
differ only by constant amounts. Conversely, the N economies diverge if the deviations
from the steady state are permanent, and hence the initial values impact in the long run
their levels.
Then, in a stochastic world, economies 1, 2, . . . , N are said to converge if, and only
if, a common trend at 7 and finite parameters µ1 , µ2 , . . . , µN exist such that:
lim Et (yn,t+i − at+i ) = µn
i→∞
(1)
for n = 1, 2, . . . , N, and ynt is the logarithm of per capita output for economy n during period t. The parameter µn determines the level of economy n’s parallel balanced
growth path. Unless all economies have identical structures, the µ’s should typically
be nonzero.
Unfortunately, the common trend is unobservable. However, under the convergence hypothesis, an estimator of its value can be obtained. Indeed, if the deviations
from the steady state are not permanent, then the cross-economy average of the per
capita income must converge to the level of the common trend:
lim Et (yt+i − at+i ) = 0
i→∞
(2)
where yt = ∑Nn=1 yn,t /N. Finally, Evans and Karras (1996) obtained the following
condition:
lim Et (yn,t+i − yt+i ) = µn
i→∞
(3)
According to this assumption, the deviations of y1,t+i , y2,t+i , . . ., yN,t+i from their
cross-economy average yt can be expected, conditional on current information to approach constant values as i approaches infinity. Note that this condition holds if, and
only if, (yn,t − y) have exhibited a much higher growth rate than the richer ones, and
hence that a catching-up is occurring. On the other hand, the convergence will be said
conditional if µn 6= 0 for some n. So, each economy has converged to its own steady
state, and only the growth rates will be equalized in the long run. Operationally, these
income convergence hypotheses require testing for the presence of a unit root in panel
7 The
series at can be thought of as the logarithm of an index of Harrod-neutral technology available
to economies 1, 2, . . . , N.
8
data. The absolute convergence is tested by panel unit root tests with no fixed individual effects, whereas the conditional convergence is tested by implementing panel unit
root tests with fixed individual effects.
3.2
Panel unit root tests
In this study, we apply two first generation tests proposed by Levin et al. (2002)
and Im et al. (2003) which are homogeneous and heterogeneous panel unit root
tests, respectively, based on the assumption of independent cross-section units. In
Levin et al. (2002), the alternative hypothesis is that no series contains a unit root
(all are stationary) while in Im et al. (2003) the alternative allows unit roots for
some (but not all) of the series8 . However, the cross-unit independence assumption
of the first generation tests is quite restrictive in many empirical applications and
can lead to severe size distortions (Banerjee et al., 2005; Breitung and Das, 2008).
Therefore, we also consider a second generation unit root tests that allow cross-unit
dependencies with the tests developed by Bai and Ng (2004). The simplest way
consists in using a factor structure model. The idea is to shift data into two unobserved
components: one with the characteristic that is cross-sectionally correlated and one
with the characteristic that is largely unit specific. Thus, the testing procedure consists
in two steps: in a first one, data are de-factored, and in a second step, panel unit root
test statistics based on de-factored data and/or common factors are then proposed. The
issue is to know if this factor structure allows obtaining clear cut conclusions about
stationarity of macroeconomic variables9 .
3.2.1 Levin, Lin and Chu (2002) test
One of the most popular first generation unit root test is undoubtedly the test proposed
by Levin, Lin and Chu (2002) [LLC]. The model with individual effects and no time
trends, in which the coefficient of the lagged dependent variable is restricted to be
homogenous across all units of the panel, is defined as
8 See
Hlouskova and Wagner (2006) for a discussion on the performance of first generation panel unit
root tests.
9 See Banerjee (1999), Baltagi and Kao (2000), Choi (2006), Hurlin (2008) and Breitung and Pesaran
(2008) for a survey on panel unit root tests. See also Gengenbach et al. (2006) and de Silva et al. (2009)
for an investigation on the properties of the second generation panel unit root tests.
9
pi
∆yit = αi + ρi yi,t−1 + ∑ βi,z ∆yi,t−1 + εit
(4)
z=1
for i = 1, . . . , N and t = 1, . . . , T . The errors εit ∼ i.i.d. (0; σ2εi ) are assumed to be
independent across the units of the sample. In this model, LLC are interested in
testing the null hypothesis H0 : ρ = 0 against the alternative hypothesis H1 : ρ = ρi < 0
for all i = 1, . . . , N, with auxiliary assumptions about the individual effects (αi = 0
for all i = 1, . . . , N under H0 ). This restrictive alternative hypothesis implies that the
autoregressive parameters are identical across the panel.
The LLC test is based on the following adjusted t-statistic
tρ∗
tρ
= ∗ − NT ŜN
σT
Ã
σ̂ρ̂
σ̂2ε̂
!µ
µ∗T
σ∗T
¶
(5)
where tρ is the standard t-statistic based on the pooled estimator ρ̂, where the mean
adjustment µ∗T and standard deviation adjustment σ∗T are simulated by LLC for various
sample sizes T . The adjustment term is also function of the average of individual
ratios of long-run to short-run variances, ŜN = (1/N) ∑Ni=1 (σ̂yi /σ̂εi ), where σ̂yi denotes
a kernel estimator of the long-run variance for the country i. LLC suggest using a
Bartlett kernel function and a homogeneous truncation lag parameter given by the
simple formula K̄ = 3.21T 1/3 . They demonstrate that, under the non stationary null
hypothesis, the adjusted t-statistic tρ∗ converges to a standard normal distribution.
3.2.2 Im, Pesaran and Shin (2003) test
Im, Pesaran and Shin (2003) [IPS] propose heterogeneous panel unit root tests based
on the cross-sectional independence assumption. The model with individual effects
and no time trend is given as
pi
∆yit = αi + ρi yi,t−1 + ∑ βi,z ∆yi,t−1 + εit
(6)
z=1
The null hypothesis is defined as H0 : ρi = 0 for all i = 1, . . . , N and the alternative is
H1 : ρi < 0 for i = 1, . . . , N1 and ρi = 0 for all i = N1 + 1, . . . , N, with 0 < N1 ≤ N. The
alternative allows unit roots for some (but not all) of the individuals. In this context, the
IPS test is based on the (augmented) Dickey-Fuller statistics averaged across groups.
Let tiT (ρi , βi ) with βi = (βi,1 , . . . , βi,ρi ) denote the t-statistics for testing unit root in the
i-th country. The IPS statistic is then defined as
10
tbarNT =
1 N
∑ tiT (ρi , βi )
N i=1
(7)
Under the assumption of cross-sectional independence, this statistic is shown to
sequentially converge to a normal distribution.
IPS propose two corresponding
standardized tbar statistics. The first one, denoted Ztbar , is based on the asymptotic
moments of the Dickey Fuller distribution. The second standardized statistic, denoted
Wtbar , is based on the means and variances of tiT (ρi , 0) evaluated by simulations under
the null ρi = 0. Although the tests Ztbar and Wtbar are asymptotically equivalent,
simulations show that the Wtbar statistic, which explicitly takes into account the
underlying ADF orders in computing the mean and the variance adjustment factors,
performs much better in small samples. For each country, the values of the mean and
variance used in the standardization of Wtbar are taken from the IPS simulations (Im,
Pesaran and Shin, 2003) for the time length T and the corresponding individual lag
order pi . Individual ADF lag orders are optimally chosen according to the general-tospecific (GS) procedure of Hall (1994) with a maximum lag length set to 410 .
3.2.3 Bai and Ng (2004) test
The unit root tests by Bai and Ng (2004) [BN] provide a complete procedure to test
the degree of integration of series. They decompose a series yit as a sum of three
components: a deterministic one, a common component expressed as a factor structure
and an error that is largely idiosyncratic. The process yit is non-stationary if one
or more of the common factors are non-stationary, or the idiosyncratic error is nonstationary, or both. Instead of testing for the presence of a unit root directly in yit , BN
propose to test the common factors and the idiosyncratic components separately. Let
us consider a model with individual effects and no time trend
yit = αi + λ′i Ft + εit
(8)
where Ft is a r × 1 vector of common factors and λi is a vector of factor loadings.
Among the r common factors, we allow r0 stationary factors and r1 stochastic common
trends with r0 + r1 = r. The corresponding model in first differences is
∆yit = λ′i ft + zit
10 Similar
(9)
results have been obtained when individual lag lengths are chosen by information criteria
(AIC or BIC).
11
where zit = ∆εit and ft = ∆Ft with E( ft ) = 0. The common factors in ∆yit are
estimated by the principal component method. Let us denote fˆt these estimates,
λ̂i the corresponding loading factors and ẑit the estimated residuals. BN propose a
differencing and re-cumulating estimation procedure which is based on the cumulated
variables
t
F̂mt =
t
∑ fˆms
ε̂it =
s=2
∑ ẑms
(10)
s=2
for m = 1, . . . , r and i = 1, . . . , N. Then, they test the unit root hypothesis in the
idiosyncratic component εit and in the common factors Ft with the estimated variables
F̂mt and ε̂it .
To test the non-stationarity of idiosyncratic components ε̂it (the de-factored
estimated components), BN suggest pooled individual ADF t-statistics from a Fisher’s
type statistic, denoted Pε̂c , rather than individual ADF t-statistics ADFcε̂ (i) in order to
improve the power of the test (BN, 2004).
To test the non-stationarity of the common factors F̂mt , BN consider a ADF t-statistic,
denoted ADFcF̂ (i), when there is only one common factor among the N variables
(r = 1). The number of common factors is estimated according to IC2 or BIC3 criteria
(see Bai and Ng, 2002) with a maximum number of factor equal to 511 .
4
Empirical analysis
4.1 The data
The data of the study consists of annual real per capita GDP data from Maddison
(2007) database for 20 COMESA economies in common 1990 Geary-Khamis PPPadjusted dollars, and spans from 1960 to 2003. Note that these data are expressed in
common 1990 Geary-Khamis PPP-adjusted dollars, which correct for the differences
in prices of commodities across countries12 .The countries represented are Angola, Burundi, Comoros, Democratic Republic of Congo, Djibouti, Egypt, Eritrea13 , Ethiopia,
Kenya, Libya, Madagascar, Malawi, Mauritius, Rwanda, Seychelles, Sudan, Swaziland, Tanzania, Uganda, Zambia, Zimbabwe. Note that all variables are expressed
in logs. Moreover, output differentials are defined with respect to the corresponding
11 BN
(2004) also consider the case when there are more than one common factors (r > 1) from a
sequential procedure. In our study, we find only one common factor.
12 See Maddison (2003) for a discussion on the Geary-Khamis approach.
13 Ethiopia and Eritrea are added into a same item.
12
panel average.
In addition, we check if there are some exogenous convergent clubs in the
COMESA by analyzing some groups of COMESA countries. Country groups are
shaped by reference to three criteria (Table 1):
(i) the degree of economic development: The importance of economic development
(human capital, health, infrastructure, . . . ) have been demonstrated since a long
time ago (Gillis et al., 1987). Recently, the New Growth Theory insisted on
the crucial impact of the initial development conditions for economic growth
and convergence. For that purpose, we first focused on the classification of
the United Nations Development Program based on the Human Development
Indicator [HDI]. Nevertheless, we fixed a threshold value of 0.6 so that we
have two groups: the High/Moderate Human Development Indicator [HMHDI]
group and the relatively Low Human Development Indicator [LHDI]. We also
retained the concept of Less Developed Countries [LDC] established by the
United Nation Conference on Trade and Development14 .
(ii) the membership to another regional trade arrangement: On the whole, all
COMESA countries belong to at least one another African trade arrangement.
This can have a significant influence on the various convergence processes in
the extent that some trade agreements are more integrated than others. More
precisely, five such organizations are concerned, namely the Indian Ocean
Commission [IOC], the East African Community [EAC], the Southern African
Development Community [SADC], the Economic Community of Central
African States [ECCAS], the Arab Maghreb Union [AMU], Intergovernmental
Authority for Development [IGAD] and the Cross-Border Initiative [CBI]15 .
14 Note
that a country is classified as a Least Developed Country (LDC) if it meets three criteria based
on: (i) low-income (three-year average GNI per capita of less than US $750, which must exceed $900 to
leave the list), (ii) human resource weakness (based on indicators of nutrition, health, education and adult
literacy) and (iii) economic vulnerability (based on instability of agricultural production, instability of
exports of goods and services, economic importance of non-traditional activities, merchandise export
concentration, and handicap of economic smallness, and the percentage of population displaced by
natural disasters).
15 The CBI was established in 1992 and consists of 14 countries (Burundi, Comoros, Kenya,
Madagascar, Malawi, Mauritius, Namibia, Rwanda, Seychelles, Swaziland, Tanzania, Uganda, Zambia,
Zimbabwe). The SADC was established in 1992 and consists of 10 countries (Angola, Botswana,
Lesotho, Malawi, Mozambique, South Africa, Swaziland, Tanzania, Zambia, Zimbabwe). The ECCAS
was created in 1985 and consists of 10 countries (Angola, Burundi, Cameroon, Central African Republic,
13
(iii) the importance of oil in the production and the export structures: Most countries
belonging to COMESA have a poor diversified export base. Some of them
strongly depend on oil resources. One more time, we can build two groups
from this criterion: the oil countries group, that is to say those which belong to
the African Petroleum Producers Association [APPA] and the non-oil countries
group [Non-APPA].
Table 1: COMESA’s countries shaped following the three criteria.
Regional integration agreement
Country
COMESA
Angola, Burundi, Comoros, Democratic Republic of Congo, Djibouti, Egypt,
Eritrea, Ethiopia, Kenya, Libya, Madagascar, Malawi, Mauritius, Rwanda,
Seychelles, Sudan, Swaziland, Tanzania, Uganda, Zambia, Zimbabwe
IOC
Comoros, Madagascar, Mauritius, Seychelles
EAC
Kenya, Tanzania, Uganda
SADC
Angola, Malawi, Swaziland, Tanzania, Zambia, Zimbabwe,
ECCAS
Angola, Burundi, D.R. Congo
AMU
Egypt, Libya, Sudan, Djibouti, Comoros
IGAD
Djibouti, Eritrea, Ethiopia, Kenya, Sudan, Uganda
CBI
Burundi, Comoros, Kenya, Madagascar, Malawi, Mauritius, Rwanda,
Mauritius, D.R. Congo, Madagascar
Seychelles, Swaziland, Tanzania, Uganda, Zambia, Zimbabwe
Economic development criterion
Country
HMIDH
Egypt, Libya, Mauritius, Seychelles
LHDI
Angola, Burundi, D.R. Congo, Comoros, Djibouti, Eritrea, Ethiopia, Kenya,
Madagascar, Malawi, Rwanda, Sudan, Swaziland, Tanzania, Uganda,
Zambia, Zimbabwe
LDC
Angola, Burundi, D.R. Congo, Comoros, Djibouti, Eritrea, Ethiopia,
Madagascar, Malawi, Rwanda, Sudan, Tanzania, Uganda, Zambia
Economic structure criterion
Country
APPA
Angola, D.R. Congo, Egypt, Libya, Sudan
Non-APPA
Burundi, Comoros, Djibouti, Eritrea, Ethiopia, Kenya, Madagascar, Malawi,
Mauritius, Rwanda, Seychelles, Swaziland, Tanzania, Uganda, Zambia, Zimbabwe
Chad, Congo, Equatorial Guinea, Gabon, São Tomé and Príncipe, and D.R. Congo). The IOC was
established in 1984 and consists of 5 countries (Comoros, France (La Réunion), Madagascar, Mauritius
et Seychelles). The EAC was created in 1999 and consists of 3 countries (Kenya, Uganda and Tanzania)
with Burundi and Rwanda joining in 2007. The IGAD was established in 1986 and consists of 7 countries
(Djibouti, Ethiopia, Eritrea, Kenya, Somalia, Sudan and Uganda).
14
4.2
Empirical results
The adopted strategy to test for income convergence is straightforward. Firstly, for
each group, we apply panel unit root tests with no fixed individual effects in order
to check if an absolute convergence process is present in the sample considered.
Secondly, for the groups where the null of unit root can not be rejected, the same
panel unit root tests but with fixed individual effects are implemented to pin downs a
possible conditional convergence dynamics. Finally, if the unit root hypothesis always
holds, then we consider that the group is characterized by stochastic divergence. Table
2 reports for the whole COMESA the panel unit root tests with no individual effects
suggested by Levin et al. (2002) [LLC1 , tρ∗ ] and with individual effects by Levin et al.
(2002) [LLC2 , tρ∗ ], Im et al. (2003) [IPS, tbarNT ] and Bai and Ng (2004) [BNc and
BNi for common factors (ADFcF̂ ) and idiosyncratic shocks (Pε̂c ), respectively.]. Table 3
displays the outcomes resulting from the same tests but for the different convergence
clubs presented in Table 1.
Table 2: Panel unit root tests for absolute and conditional convergence: Results for the
whole COMESA.
References
The COMESA average
The African average
The world average
∗
and
∗∗
LLC1
LLC2
3.82
1.43
(0.99)
(0.92)
3.38
2.31
(0.99)
(0.99)
8.88
3.46
(1.00)
(0.99)
IPS
BNc
BNi
3.04
−1.55
32.48
(0.99)
2.62
(0.99)
6.22
(1.00)
(0.50)
(0.80)
−0.34
30.90
(0.92)
(0.85)
1.20
41.13
(0.99)
(0.42)
Significant at the 5% and 10% level, respectively. LLC1 and LLC2 denote the Levin, Liu and Chin (2002)
panel unit root test with no individual effects and with individual effects respectively. IPS denotes the Im, Pesaran and
Shin (2003) unit root test with individual unit root processes. BNc and BNi denote the Bai and Ng (2004) secondgeneration unit root test for common factors (ADFcF̂ ) and idiosyncratic shocks (Pε̂c ), respectively. Note that all these
three tests are done with individual effects.
4.2.1 The regional integration criterion
Implementing the unit root tests on the COMESA16 and the memberships to another
African regional trade agreement (IOC, EAC, SADC, ECCAS, AMU, IGAD, CBI)
gives, with few exceptions, strong support for rejecting the hypotheses of absolute
16 For
the whole COMESA, Table 2 shows that the finding of no absolute and conditional convergence
holds even if other income references (an African average and a world average) are used.
15
Table 3: Panel unit root tests for absolute and conditional convergence: Results for the
convergence clubs.
Groups
LLC1
LLC2
IPS
BNc
BNi
Grouping by the regional integration criterion
ECCAS
0.61
(0.73)
SADC
4.82
EAC
AMU
CBI
4.78
2.71
(1.00)
5.96
−1.25
2.96
(0.65)
(0.94)
(1.00)
(1.00)
(1.00)
(0.99)
0.83
15.29
−0.79
1.25
1.63
−1.42
15.69∗
(0.22)
(0.89)
(0.95)
−1.26
−0.69
−0.16
(0.11)
(0.25)
(0.44)
4.68
IGAD
2.22
(0.99)
3.07
5.61
(0.50)
(0.57)
(0.02)
2.77
−1.37
(0.99)
(0.56)
(1.00)
(1.00)
(1.00)
(0.98)
0.34
32.26
−2.24∗
–
–
–
–
(0.19)
(0.01)
IOC
4.57
(1.00)
3.27
(1.00)
4.00
(1.00)
0.48
(0.99)
8.18
(0.41)
Grouping by the economic structure criterion
APPA
NON-APPA
−0.15
−0.61
(0.44)
(0.27)
4.34
(1.00)
3.49
(0.99)
0.29
(0.61)
5.83
(1.00)
−1.29
2.56
(0.63)
(0.98)
0.37
37.45
(0.98)
(0.16)
Grouping by the economic development criterion
LHDI
HMHDI
−0.32
(0.38)
−1.42∗∗
1.42
2.70
(0.92)
(0.99)
–
–
−1.67
27.83
(0.45)
(0.68)
–
–
–
–
–
–
(0.08)
LDCs
−2.45∗
(0.01)
∗
and
∗∗
Significant at the 5% and 10% level, respectively. LLC1 and LLC2 denote the Levin, Liu and Chin (2002)
panel unit root test with no individual effects and with individual effects respectively. IPS denotes the Im, Pesaran and
Shin (2003) unit root test with individual unit root processes. BNc and BNi denote the Bai and Ng (2004) secondgeneration unit root test for common factors (ADFcF̂ ) and idiosyncratic shocks (Pε̂c ), respectively. Note that all these
three tests are done with individual effects.
and conditional convergence. Indeed, we cannot reject the null hypothesis of nonstationarity for all trade arrangements whatever the specification retained, i.e. with
or without a fixed individual effect, except for the African trade arrangements IGAD
and EAC. For the IGAD group the panel unit root test with no individual effects LLC
does not reject the null hypothesis of a unit root at the 5% level, implying an absolute convergence across Djibouti, Eritrea-Ethiopia, Kenya, Sudan and Uganda, i.e. the
level of per capita income of the poor countries (Burundi and Eritrea-Ethiopia) in this
16
group catch-up with the one of the rich ones (Kenya). For the EAC group the panel
unit root test with individual effects BNi does not reject the null hypothesis at the 5%
level, suggesting a conditional convergence across Kenya, Tanzania and Uganda, i.e.
each economy has converged to its own steady state, and only the growth rates will be
equalized in the long run. This property of non-stationarity is due to the behavior of
the common component (growth trends).
However, note that this finding of no convergence process for the trade
arrangement criterion does not reveal that regional integration is not an efficient
strategy to make developing countries to converge. In our point of view, these results
just tell us that the ongoing process of integration is not adapted in this part of Africa.
In accordance with the so-called Spaghetti Bowl effect of Bhagwati et al. (1998), the
high number of trade agreements in Eastern and Southern Africa contributes to this
bad performance in terms of income convergence.
4.2.2 The economic structure criterion
If we use the economic structure criterion, no clubs convergence appears. The null of
a unit root is not rejected by all the tests both for the absolute and conditional convergence hypothesis whatever the group considered (APPA, Non-APPA). Moreover,
taking into account the presence of cross-sectional dependence does not change the
results. The reject of the convergence for these two groups is not very surprising. The
discrimination by the oil criterion is not sufficient to constitute homogeneous groups
in the case of the COMESA. Several members reveal a production structure more diversified as for instance Egypt, Mauritius or Seychelles.
4.2.3 The economic development criterion
Finally, the grouping by the economic development criterion gives the more interesting
findings. Two out of three groups are associated with an absolute income convergence
trend. In effect, the null hypothesis of a unit root can be rejected at the 5% and 10%
level for the LDC and HMIDH groups, respectively, from the panel unit root test with
no individual effects LLC. This result implies that the level of per capita income of the
poor countries in these groups catch-up with the one of the rich ones. Concerning the
last one, the LHDI group, a divergent process seems to characterize the data, i.e. the
group do not converge both in level and in rate. That is not very surprising because
17
of the strong economic development disparities which are still present in this group.
Indeed, some countries as Zimbabwe, Kenya or Swaziland reveal HDI performances
close to the upper limit of 0.6. Although their economic development levels stay relatively low, they do undoubtedly better than the 14 other countries.
Thus, our work allows us to strongly support the theoretical insight. Economic development is crucial for improving the growth performances of an economy. This conjecture is more evident for the COMESA. Countries with good economic development
conditions (Mauritius, Seychelles, Libya, Egypt) show a catching up process towards
a high income average. But, countries with bad economic development conditions, i.e.
sixteen out of twenty economies, converge towards a low income average. All in all,
we can conclude from this study that there is an income convergence process towards
the bottom within the COMESA. Indeed, except for four countries, all the members of
this regional agreement are locked into the poverty trap.
5
Conclusion
In this paper, we proposed to detect the possibility of stochastic convergence of real
per capita GDP for a set of Eastern and Southern African countries, all members of the
COMESA’s trade agreement. Using the panel unit-root tests developed by Levin et al.
(2002), Im et al. (2003) and Bai and Ng (2004), our results rejected the presence of
stochastic convergence for the whole COMESA.
However, in the extent that most COMESA countries are largely heterogeneous,
we tried to put forward the potential existence of convergence clubs within the trade
agreement by three criteria, namely (i) the membership to another regional agreement,
(ii) the economic structure (dependence from oil production) and (iii) the degree of
global economic development. Two main findings emerged from the results. Firstly,
no evidence of stochastic absolute and conditional income convergence holds for all
the groups belonging to the regional trade agreement and the economic structure
criteria. Concerning the former criterion, contrary to the conceptual conclusion of
Venables (2003) about South-South integration, the lack of convergence in our case
does not imply that regional integration does not stimulate the setting up of a catching
up process. Actually, in our point of view, this bad performance results from the
so-called « Spaghetti Bowl » effect of Bhagwati et al. (1998). Thus, this region
needs a strategy based on a rationalization of the number of trade agreements before
18
deepening the trade and financial relations between the different economies. Secondly,
the testing procedures highlighted strong support for absolute income convergence
for two groups (HMIDH, LDC) belonging to the economic development criterion.
This result led us to conclude that a convergence process towards the bottom is at
work for the COMESA members, except for the most four developed countries, that
is Mauritius, Seychelles, Libya and Egypt. This result corroborates the findings of
the New Growth Theories in the extent that initial economic development conditions
determine the long-run economic growth processes. A related outcome is the necessary
intervention of both local governments and international institutions to create a climate
of sustainable development and get these under-development economies out of the
poverty trap.
Further research should investigate the convergence in the COMESA by using
panel unit root tests taking into account a break as in, e.g., Carrion-i-Silvestre et al.
(2005).
19
References
[1] Bai J., Ng S. (2002), "Determining the number of factors in approximate factor
models", Econometrica, 70, 191-221.
[2] Bai J., Ng S. (2004), "A PANIC attack on unit roots and cointegration",
Econometrica, 72, 1127-1178.
[3] Baltagi B.H., Kao C. (2000), "Nonstationary panels, cointegration in panels
and dynamic panels: A survey", in Advances in Econometrics, Elsevier, North
Holland.
[4] Banerjee A. (1999), "Panel data unit root and cointegration: An Overview",
Oxford Bulletin of Economics and Statistics, Special Issue, 607-629.
[5] Banerjee A., Marcellino M., Osbat C. (2005), "Testing for PPP: Should we use
panel methods?", Empirical Economics, 30, 77-91.
[6] Banerjee A., Lumsdaine R.L., Stock J.H. (1992), "Recursive and sequential test
of the unit-root and trend-break hypothesis: Theory and international evidence",
Journal of Business and Economics Statistics, 10, 271-287.
[7] Barro R.J., Sala-i-Martin X. (1991), "Convergence across states and regions",
Brookings Papers on Economic Activity, 1, 107-158.
[8] Barro R.J., Sala-i-Martin X. (1992), "Convergence", Journal of Political
Economy, 100, 223-251.
[9] Baumol W. J. (1986), "Productivity growth, convergence and welfare: What the
long run data show", American Economic Review, 76, 1072-1085.
[10] Bernard A., Durlauf S. (1995) "Convergence in international output", Journal of
Applied Econometrics, 10, 97-108.
[11] Bernard A., Durlauf S. (1996) "Interpreting tests of the convergence hypothesis",
Journal of Econometrics, 71, 161-173.
[12] Bertocchi G., Canova F. (2002), "Did colonization matter for growth? An
empirical exploration into the historical causes of Africa’s underdevelopement",
European Economic Review, 46, 1851-1873.
[13] Bhagwati J., Greenaway D., Panagariya A. (1998), "Trading preferentially:
Theory and policy", The Economic Journal, 108, 1128-1148, July.
20
[14] Block S.A. (2001), "Does Africa grow differently?", Journal of Development
Economics, 65, 443-467.
[15] Bloom D.E., Sachs J.D. (1998), "Geography, demography, and economic growth
in Africa", Brookings Papers on Economics, 207-295.
[16] Breitung J., Das S. (2008), "Testing for unit roots in panels with a factor
structure", Econometric Theory, 24, 88-108.
[17] Breitung J., Pesaran M. (2008), "Unit roots and cointegration in panels", in
Matyas L. and Sevestre P. (eds.), The Econometrics of Panel Data, New Jersey:
Springer-Verlag.
[18] Carmignani F. (2006), "The road to regional integration in Africa: Macroeconomic convergence and performance in COMESA", Journal of African
Economies, 15(2), 212-250, June.
[19] Carmignani F. (2007), "A note on income converge effects in regional integration
agreements", Economics Letters, 94, 361-366.
[20] Carrion-i-Silvestre J-L., del Barrio-Castro T., Lopez-Bazo E. (2005), "Breaking
the panels: An application to the GDP per capita", Econometrics Journal, 8,
159-175.
[21] Choi I. (2001), "Unit root tests for panel data", Journal of International Money
and Finance, 20, 249-272.
[22] Choi I. (2006), "Nonstationary panels", in Mills T.C., Patterson K. (eds.),
Palgrave Handbook of Econometrics: Econometric Theory, Palgrave Macmillan.
[23] Collier P., Gunning J.W. (1999), "Why has Africa grown slowly?", Journal of
Economic Perspectives, 13, 3-22.
[24] Cuñado J., Pérez de Gracia F. (2006), "Real convergence in Africa in the secondhalf of the 20th century", Journal of Economics and Business, 58, 153-167.
[25] de Silva S., Hadri K., Tremayne A.R. (2009), "Panel unit root tests in the presence
of cross-sectional dependence: finite sample performance and an application",
Econometrics Journal, 12, 340-366.
[26] Easterly W., Levine R. (1997), "Africa’s growth tragedy: Policies and ethnic
divisions", Quarterly Journal of Economics, 112, 1203-1250.
21
[27] Evans P. (1996), "Using cross-country variances to evaluate growth theories",
Journal of Economic Dynamics and Control, 20, 1027-1049.
[28] Evans P., Karras, G. (1996), "Convergence revisited", Journal of Monetary
Economics, 37, 249-265.
[29] Gengenbach C., Palm F.C., Urbain J-P. (2006), "Panel unit root tests in the
presence of cross-sectional dependencies: Comparison and implications for
modelling", METEOR Research Memorandum 04039, Maastricht University.
[30] Guetat I., Serranito F.. (2007), "Income convergence within the MENA countries:
A panel unit root approach", The Quaterly Review of Economics and Finance, 46,
685-706.
[31] Haldrup N., Jansson M. (2006), "Improving size and power in unit root
testing", in Mills T.C., Patterson K. (eds.), Palgrave Handbook of Econometrics:
Econometric Theory, Palgrave Macmillan.
[32] Hall A., (1994), "Testing for a unit root in time series with pretest data-based
model selection", Journal of Business and Economic Statistics, 12, 461-70.
[33] Hlouskova J., Wagner, M. (2006), "The performance of panel unit root and
stationarity tests: Results from a large scale simulation study", Econometric
Reviews, 25, 85-116.
[34] Hurlin C., "What would Nelson and Plosser find had they used panel unit root
tests?", Applied Economics, forthcoming.
[35] Im K.S., Lee J., Tieslau M. (2007) "Panel LM unit-root tests with level shifts",
Oxford Bulletin of Economics and Statistics, 67, 393-419.
[36] Im K.S., Pesaran M.H., Shin Y. (2003), "Testing for unit roots in heterogeneous
panels", Journal of Econometrics, 115, 53-74.
[37] Islam N. (2003), "What have we learnt from the convergence debate?", Journal
of Economic Survey, 17, 309-362.
[38] Levin A., Lin C.F., Chu. C.S.J. (2002), "Unit root test in panel data: Asymptotic
and finite sample properties", Journal of Econometrics, 108, 1-24.
[39] Lima Marcos A.M., Resende M. (2007), "Convergence of per capita GDP in
Brazil: An empirical note", Applied Economics Letters, 14, 333-335.
22
[40] MacKinnon J.G. (1991), Critical values for cointegration tests. In: Engle,
R.F., Granger, C.W. (Eds.), Long-Run Economic Relationship: Readings in
Cointegration. Oxford Press, Oxford.
[41] MacKinnon J.G. (1994), "Approximate asymptotic distribution functions for
unit-root and cointegration", Journal of Business and Economic Statistics, 12,
167-176.
[42] Maddala G.S., Wu S. (1999), "A comparative study of unit root tests with panel
data and a new simple test", Oxford Bulletin of Economics and Statistics, 61,
631-652.
[43] Maddison A. (2003), The World Economy: A Millennial Perspective, Development Center for the OECD, Paris.
[44] McCoskey S.K. (2002), "Convergence in Sub-Saharan Africa: A nonstationary
panel data approach", Applied Economics, 34, 819-829.
[45] McCoskey S.K., Kao C. (1998), "A residual-based test of the null of cointegration
in panel data", Econometric Reviews, 17, 57-84.
[46] Paap R., Franses P.H., van Dijk D. (2005), "Does Africa grow slower than
Asia, Latin America and the Middle East? Evidence from a new data-based
classification method", Journal of Development Economics, 77, 553-570.
[47] Quah D. (1993), "Galton’s fallacy and tests of the convergence hypothesis",
Scandinavian Journal of Economics, 95, 427-443.
[48] Quah D. (1996), "Empirics for economic growth and convergence", European
Economic Review, 40, 1353-1375.
[49] Romer P. (1986), "Increasing returns and long-run growth", Journal of Political
Economy, 94, 1002-1037.
[50] Romer P. (1990), "Endogenous technical change", Journal of Political Economy,
98(5), S71-S102.
[51] Solow R. (1956), "A contribution to the theory of economic growth", Quarterly
Journal of Economics, 70, 65-94.
[52] Venables J. (2003), "Winners and losers from regional integration agreements",
Economic Journal, 113, 747-761.
23